Improving Parallel Performance of Temporally Relevant Top-K Spatial Keyword Search

Abstract

With the rapid growth of geotagged documents, top-k spatial keyword search queries (TkSKQ) have attracted a lot of attention and a number of spatio-textual indexes have been proposed. While some indexes support real-time updates over continuously generated documents, they do not support queries that simultaneously consider temporal relevance, textual similarity ranking and spatial location. Existing indexes also have limited capability to exploit parallelism. To address these issues, we introduce a novel parallel index, called Pastri (PArallel Spatio-Textual adaptive Ranking-based Index), which can be incrementally updated based on live spatio-textual document streams. Pastri uses a dynamic ranking scheme to retrieve the top-k objects that are most temporally relevant at the time of a query execution. We have built a system in which we integrate Pastri along with a persistent document store and several thread pools to exploit parallelism at various levels. Experimental evaluation demonstrates that our system can support high document update throughput and low latency with TkSKQ queries.